It was a lighter week for insurance software news, but three stories stood out: a regulator questioning carriers about vendor AI, two agentic AI launches with the same built-in limits, and a quarterly report that tests AI-first distribution. Here is what happened and why we think each one matters.
1. California asks insurers how they govern AI, including their vendors’ AI
What happened: On November 16, the California Department of Insurance opened a survey on insurers’ use of AI. It covers governance, reliance on third-party vendors and operational effects. Responses are due December 30, and a virtual public workshop follows on December 15. (Insurance Journal)
Our take: The workshop lists insurer responsibility for vendor technology as its own topic. For carriers whose AI sits mostly inside vendor platforms, documentation and audit rights are becoming a compliance issue. Read more in California just said it: your vendor’s AI is your AI.
What to check in your own stack:
- Do you have a complete inventory of AI features, including the ones built into vendor products?
- Could you produce vendor testing evidence and model documentation on request?
- Do your contracts include audit rights and notice of model changes?
2. Two agentic AI launches, one design rule: the AI doesn’t make the final call
What happened: On November 13, Optalitix added AI agents to its quoting platform. The agents work on top of existing pricing models and approval steps and connect to ChatGPT, Claude and Copilot over MCP (FinTech Global). The same day, OutSystems launched an agentic first-notice-of-loss product for auto claims, where adjusters keep coverage, fault and payment decisions (OutSystems press release via Insurance Journal).
Our take: Both vendors let agents gather, check and prepare, but leave decisions to people or approved models. Expect that “bounded agent” design to become the norm in regulated insurance workflows. Read more in Agents that don’t decide.
Questions to ask before buying:
- Which actions can the agent take without approval?
- Can you replay what the agent saw and did for any single case?
- If assistants connect over MCP, how are data access rights enforced?
3. Lemonade’s Q3: fast growth, a wider loss and a Q4 test
What happened: On November 10, Lemonade reported Q3 revenue up 73% to $336.1 million and in-force premium up 34% to $1.55 billion. Its net loss widened to about $43.7 million. Lemonade still expects its first adjusted EBITDA-positive quarter in Q4. (Insurance Journal)
Our take: The numbers say more about distribution than about AI. Direct digital channels can scale quickly, but customer acquisition is still expensive. Q4 will show whether the model can grow and turn profitable. Read more in Lemonade’s Q3 in five numbers.
What this means in practice:
- Expect AI savings in the expense ratio first, and in the loss ratio later.
- Direct distribution economics depend on selling several products to each customer.
- Watch loss ratios in new car insurance states as pricing data builds up.
Also from the week: GuideOne Insurance moved its commercial lines core systems to Guidewire Cloud in ten months. We looked at why that matters in A 10-month cloud move shows why core systems are now an AI project.